Methods for explainable artificial intelligence
Sajid Ali · Institution of Engineering and Technology eBooks · 2023
As AI models are becoming increasingly regulated by governments, it has become crucial to provide explanations for their decisions. The emergence of XAI has helped us to better understand AI systems and move towards models that can offer human-friendly explanations. However, it remains unclear whether the growing range of XAI methodologies and tools is enough to provide practical support in the risky scenarios that regulatory stakeholders are concerned about. For instance, can an intelligent model be used for a medical diagnosis simply because of the availability of score-CAM or GradCAM? The answer is a resounding "NO" because there are no established risk-aware scenarios that can guide the research community on the requirements for implementing XAI-supported AI models in real-world contexts. Therefore, society needs approaches that recognize XAI tools as necessary but insufficient steps toward assessing the trustworthiness of AI-based systems for specific tasks.